intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2025-12-30 01:35:20 +08:00
Code Issues Packages Projects Releases Wiki Activity

feature: add getter of walker inline data offset

Browse Source 
Related-To: NEO-10381

Signed-off-by: Zbigniew Zdanowicz <zbigniew.zdanowicz@intel.com>
This commit is contained in:
Zbigniew Zdanowicz
2024-03-27 20:10:32 +00:00
committed by Compute-Runtime-Automation
parent 3b1e6b61e4
commit 489ef2a310
5 changed files with 35 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
shared/source/command_container/command_encoder.h
View File

@@ -189,6 +189,8 @@ struct EncodeDispatchKernel {
template <bool isHeapless>
static void setScratchAddress(uint64_t &scratchAddress, uint32_t requiredScratchSlot0Size, uint32_t requiredScratchSlot1Size, IndirectHeap *ssh, CommandStreamReceiver &csr);
static size_t getInlineDataOffset(EncodeDispatchKernelArgs &args);
};
template <typename GfxFamily>

5
shared/source/command_container/command_encoder_bdw_and_later.inl
View File

@@ -630,6 +630,11 @@ size_t EncodeDispatchKernel<Family>::additionalSizeRequiredDsh(uint32_t iddCount
return iddCount * sizeof(typename Family::INTERFACE_DESCRIPTOR_DATA);
}
template <typename Family>
inline size_t EncodeDispatchKernel<Family>::getInlineDataOffset(EncodeDispatchKernelArgs &args) {
return 0;
}
template <typename Family>
size_t EncodeStates<Family>::getSshHeapSize() {
return 64 * MemoryConstants::kiloByte;

20
shared/source/command_container/command_encoder_xehp_and_later.inl
View File

@@ -320,6 +320,11 @@ void EncodeDispatchKernel<Family>::encode(CommandContainer &container, EncodeDis
}
if constexpr (heaplessModeEnabled) {
auto inlineDataPointer = reinterpret_cast<char *>(walkerCmd.getInlineDataPointer());
auto indirectDataPointerAddress = kernelDescriptor.payloadMappings.implicitArgs.indirectDataPointerAddress;
auto heap = container.getIndirectHeap(HeapType::indirectObject);
auto address = heap->getHeapGpuBase() + offsetThreadData;
std::memcpy(inlineDataPointer + indirectDataPointerAddress.offset, &address, indirectDataPointerAddress.pointerSize);
auto requiredScratchSlot0Size = kernelDescriptor.kernelAttributes.perThreadScratchSize[0];
auto requiredScratchSlot1Size = kernelDescriptor.kernelAttributes.perThreadScratchSize[1];
@@ -327,18 +332,9 @@ void EncodeDispatchKernel<Family>::encode(CommandContainer &container, EncodeDis
auto ssh = container.getIndirectHeap(HeapType::surfaceState);
uint64_t scratchAddress = 0u;
EncodeDispatchKernel<Family>::template setScratchAddress<heaplessModeEnabled>(scratchAddress, requiredScratchSlot0Size, requiredScratchSlot1Size, ssh, *csr);
auto inlineDataPointer = reinterpret_cast<char *>(walkerCmd.getInlineDataPointer());
auto indirectDataPointerAddress = kernelDescriptor.payloadMappings.implicitArgs.indirectDataPointerAddress;
auto heap = container.getIndirectHeap(HeapType::indirectObject);
auto address = heap->getHeapGpuBase() + offsetThreadData;
std::memcpy(inlineDataPointer + indirectDataPointerAddress.offset, &address, indirectDataPointerAddress.pointerSize);
auto scratchPointerAddress = kernelDescriptor.payloadMappings.implicitArgs.scratchPointerAddress;
std::memcpy(inlineDataPointer + scratchPointerAddress.offset, &scratchAddress, scratchPointerAddress.pointerSize);
} else {
walkerCmd.setIndirectDataStartAddress(static_cast<uint32_t>(offsetThreadData));
walkerCmd.setIndirectDataLength(sizeThreadData);
@@ -926,6 +922,12 @@ size_t EncodeDispatchKernel<Family>::additionalSizeRequiredDsh(uint32_t iddCount
return 0u;
}
template <typename Family>
inline size_t EncodeDispatchKernel<Family>::getInlineDataOffset(EncodeDispatchKernelArgs &args) {
using DefaultWalkerType = typename Family::DefaultWalkerType;
return offsetof(DefaultWalkerType, TheStructure.Common.InlineData);
}
template <typename Family>
size_t EncodeStates<Family>::getSshHeapSize() {
return 2 * MemoryConstants::megaByte;

8
shared/test/unit_test/encoders/test_encode_dispatch_kernel.cpp
View File

@@ -1628,4 +1628,10 @@ HWTEST_F(CommandEncodeStatesTest, givenCommandContainerWhenIsKernelDispatchedFro
dispatchArgs.isKernelDispatchedFromImmediateCmdList = true;
EncodeDispatchKernel<FamilyType>::template encode<DefaultWalkerType>(*cmdContainer.get(), dispatchArgs);
EXPECT_NE(0u, cmdContainer->getHeapWithRequiredSizeAndAlignmentCalled);
}
}
HWCMDTEST_F(IGFX_GEN8_CORE, CommandEncodeStatesTest, givenEncodeDispatchKernelWhenGettingInlineDataOffsetThenReturnZero) {
EncodeDispatchKernelArgs dispatchArgs = {};
EXPECT_EQ(0u, EncodeDispatchKernel<FamilyType>::getInlineDataOffset(dispatchArgs));
}

10
shared/test/unit_test/encoders/test_encode_dispatch_kernel_xehp_and_later.cpp
View File

@@ -1557,3 +1557,13 @@ HWTEST2_F(EncodeKernelScratchProgrammingTest, givenHeaplessModeDisabledWhenSetSc
uint64_t expectedScratchAddress = 0;
EXPECT_EQ(expectedScratchAddress, scratchAddress);
}
HWTEST2_F(CommandEncodeStatesTest, givenEncodeDispatchKernelWhenGettingInlineDataOffsetThenReturnWalkerInlineOffset, IsAtLeastXeHpCore) {
using DefaultWalkerType = typename FamilyType::DefaultWalkerType;
EncodeDispatchKernelArgs dispatchArgs = {};
size_t expectedOffset = offsetof(DefaultWalkerType, TheStructure.Common.InlineData);
EXPECT_EQ(expectedOffset, EncodeDispatchKernel<FamilyType>::getInlineDataOffset(dispatchArgs));
}